Convergence Guarantees for Muon Optimizer and Muesterov Variant

Muon is proven to satisfy asymptotic convergence and linear function value convergence under a global Polyak-Lojasiewicz condition. The analysis treats the Newton-Schulz implementation as a bounded preconditioner, enabling a new Nesterov-based variant called Muesterov with identical guarantees.